Ask about customized geometry and FoV of agent

[JinJinSang]

Thanks for making this amazing project open-source for the community, really appreciate it!

I am a fresh researcher just entering this field and may ask a bit about 3 questions (hope will not bother you too much!!):

1. Am I able to customize my own geometry shape? Is the collision check of different geometry based on mesh collision check or just by calculating "distance to the surface of that geometry" ?
2. Am I able to run much more parallel environments? For example, 200 simulation environment all together to further accellerate the training? Is it more related to the VRAM or the thread of the cpu?
3. Is that easy to add more features into this simulator? For example, add limited FoV for the robot (mean can only see surrounding human within the FoV)

Sorry for bothering you about these questions. Thx first!

[Shuijing725]

1. The collision check with humans and static obstacles are based on mesh, but for walls it's based on the geometry (see code here).
2. Yes, you can change training.num_processes in crowd_nav/configs/config.py. The speed of training is definitely related CPU threads, but I haven't checked VRAM (My guess is yes since the size of VRAM determines the max number of environments).
3. Yes, you can change robot.FOV in crowd_nav/configs/config.py.